CN115731166A - High-voltage cable connector polishing defect detection method based on deep learning - Google Patents
High-voltage cable connector polishing defect detection method based on deep learning Download PDFInfo
- Publication number
- CN115731166A CN115731166A CN202211230402.1A CN202211230402A CN115731166A CN 115731166 A CN115731166 A CN 115731166A CN 202211230402 A CN202211230402 A CN 202211230402A CN 115731166 A CN115731166 A CN 115731166A
- Authority
- CN
- China
- Prior art keywords
- picture
- voltage cable
- training
- defect
- deep learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a high-voltage cable joint polishing defect detection method based on deep learning, which comprises the following steps: collecting a picture of the high-voltage cable connector after manual polishing by using an industrial camera; after preprocessing and background segmentation are carried out on the obtained picture, the position and the type of the defect are manually marked on the segmented picture by utilizing marking software; cutting the marked picture to form a data set, performing data augmentation on the data set, and dividing the data set into a training set and a verification set according to a certain proportion; inputting the training set and the verification set into a target detection network based on deep learning for training to generate a detection model; and inputting the high-voltage cable connector polishing picture to be detected into the trained detection model to obtain all defect positions and defect types in the picture. The invention utilizes a machine vision method to detect the polishing defect of the high-voltage cable joint, provides a set of complete and feasible detection method and operation flow, and improves the effectiveness and feasibility of the defect detection method.
Description
Technical Field
The invention relates to the technical field of defect identification and detection, in particular to a high-voltage cable connector polishing defect detection method based on deep learning.
Background
High voltage cables often require special connections during the laying process, wherein the cable joints need to be manually ground into a specific shape before being put into use in order to have better transmission performance and durability. However, the manual grinding method often causes many defects, such as unsmooth surface, scratches, uneven transition zone and the like, which affect the performance of the cable, and therefore, is very important for detecting the defects of the cable joint. In the past, people generally used manual detection methods to check these defects, such as visual observation or manual touch by experienced teachers, however, this method is not only time-consuming and labor-consuming, inefficient, but also difficult to form a uniform standard for subjective reasons.
The defect detection method of the high-voltage cable joint polishing belongs to the field of surface defect detection in the industrial field, the defect detection is developed to the present, the detection method can be roughly divided into a non-visual method and a visual method, the non-visual method comprises a magnetic powder detection method, a penetration detection method, an eddy current detection method, an X-ray detection method, an ultrasonic detection method and the like, the methods are often required to be contacted with a detected object or a strong electric field is applied to the detected object, when the method is used for detecting the high-voltage cable, although the cable insulation layer cannot be directly punctured, the aging of the cable can be accelerated to reduce the service life of the cable, and the long-term partial discharge phenomenon can cause the insulation to be bad and even to be punctured; the vision-based method can be divided into a traditional manual characteristic-based method and a deep learning method, and compared with a non-vision-based method, the vision-based method can realize non-electric quantity non-contact online detection of the surface defects of the cable.
At present, deep learning is widely applied to the field of machine vision defect detection, and compared with a traditional method based on manual feature extraction, the deep learning automatically extracts features by using convolution operation, and a detection model which is most suitable for a product is automatically generated by using a large amount of data training, so that the method has the advantages of robustness, accuracy, strong transportability and the like, and is widely applied to industrial detection such as PCB (printed circuit board) defect and steel surface in recent years. However, there is no mature scheme for high voltage cable joint grinding defect detection based on deep learning.
Disclosure of Invention
The invention aims to provide a high-voltage cable joint polishing defect detection method based on deep learning, which is used for detecting the polishing defect of a high-voltage cable joint by using a machine vision method, provides a set of complete and feasible detection method and operation flow from data acquisition, data manufacturing, model training to final detection, and improves the effectiveness and feasibility of the defect detection method.
In order to achieve the purpose, the invention provides the following scheme:
a high-voltage cable joint polishing defect detection method based on deep learning comprises the following steps:
s1, acquiring a picture of a high-voltage cable connector which is polished manually by using an industrial camera;
s2, after preprocessing and background segmentation are carried out on the obtained picture, the position and the type of the defect are manually marked on the segmented picture by utilizing marking software;
s3, cutting the marked picture to form a data set, performing data augmentation on the data set, and dividing the data set into a training set and a verification set according to a certain proportion;
s4, inputting the training set and the verification set into a target detection network based on deep learning for training to generate a detection model;
and S5, inputting the polished picture of the high-voltage cable joint to be detected into the trained detection model to obtain all defect positions and defect types in the picture.
Further, in step S1, utilize the industry camera to gather the high tension cable joint picture after the manual work is polished, specifically include:
the four industrial cameras are respectively positioned at four angles, namely, the upper angle, the lower angle, the left angle and the right angle and are perpendicular to the high-voltage cable to shoot, the distance between each industrial camera and the high-voltage cable is equal, the four industrial cameras are positioned on the same plane in space and are perpendicular to the side face and parallel to the top face of the high-voltage cable.
Further, in step S2, after the obtained picture is preprocessed and background-segmented, the position and the type of the defect are manually marked on the segmented picture by using marking software, which specifically includes:
s201, preprocessing the picture: correcting the phenomenon of uneven illumination of the picture by adopting an illumination uneven image self-adaptive correction algorithm based on a two-dimensional gamma function, and removing noise by adopting Gaussian filtering;
s202, performing background segmentation on the picture: removing the background by using an HSV color threshold segmentation based method, and then shrinking the width of the picture;
and S203, labeling the segmented pictures by using Labelme labeling software, and generating COCO or VOC format storage according to the selected target detection network data format.
Further, in step S3, the marked picture is cut to form a data set, the data set is augmented, and the data set is divided into a training set and a verification set according to a certain proportion, which specifically includes:
cutting the marked picture into small pictures of 800 multiplied by 800;
random image turnover, random brightness transformation and random scale scaling are carried out on the data set, and data augmentation is achieved;
according to the following steps: 3. 8:2 or 9:1, the data set is divided into a training set and a validation set.
Further, in step S4, the training set and the verification set are input to a target detection network based on deep learning to be trained, so as to generate a detection model, which specifically includes:
adopting CenterNet as a target detection network, using Resnet50 as a backbone network, and using a model trained on a Pascal VOC2012 data set as a pre-training model;
the training process comprises a freezing training stage and a thawing training stage, 50 rounds of freezing training are performed, the batch _ size is set to be 6, the initial learning rate is set to be le-3, 250 rounds of thawing training are performed, the batch _ size is set to be 2, the initial learning rate is set to be 1e-4, an Adam optimizer is adopted, the weight attenuation coefficient is 4e-4, an equal-interval adjustment learning strategy Step _ LR is adopted, the adjustment interval Step _ size is 1, and the adjustment multiplying power gamma =0.93;
in the training process, verification and model storage are carried out in each period, and the model with the highest verification accuracy is compared and stored at the same time.
Further, in step S5, inputting the polished picture of the high-voltage cable joint to be detected into the trained detection model to obtain all defect positions and defect types in the picture, which specifically includes:
carrying out illumination enhancement on a polished picture of the high-voltage cable joint to be detected, wherein the specific illumination enhancement method is an illumination unevenness image self-adaptive correction algorithm based on a two-dimensional gamma function;
carrying out background segmentation on the enhanced picture, wherein the specific background segmentation method is based on an HSV color threshold segmentation algorithm;
splitting the picture without the background into small N × N pictures, and sending the small N pictures into a trained detection model for detection;
and splicing the detected small images back to the original size in sequence, and outputting the images.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: the method for detecting the polishing defect of the high-voltage cable joint based on deep learning can solve the problems that the polishing defect of the high-voltage cable joint is difficult to detect, the efficiency is low, no unified evaluation standard exists, a non-visual method is difficult to directly carry out nondestructive detection on the polishing defect and the like.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a schematic flow chart of a high-voltage cable joint polishing defect detection method based on deep learning according to an embodiment of the invention;
FIG. 2 is a schematic diagram of a relative position structure of an industrial camera and a high voltage cable according to an embodiment of the present invention;
FIG. 3 is a flow chart of dataset preprocessing according to an embodiment of the present invention;
FIG. 4 is a color model of an embodiment of the present invention;
FIG. 5 is a flow chart of data set processing according to an embodiment of the present invention;
FIG. 6 is a flowchart of image detection according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a high-voltage cable joint polishing defect detection method based on deep learning, which is used for detecting the polishing defect of a high-voltage cable joint by using a machine vision method, provides a set of complete and feasible detection method and operation flow from data acquisition, data manufacturing, model training to final detection, and improves the effectiveness and feasibility of the defect detection method.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in fig. 1, the method for detecting grinding defects of a high-voltage cable joint based on deep learning provided by the invention comprises the following steps:
s1, acquiring a picture of a high-voltage cable connector which is polished manually by using an industrial camera;
s2, after preprocessing and background segmentation are carried out on the obtained picture, the position and the type of the defect are manually marked on the segmented picture by utilizing marking software;
s3, cutting the marked picture to form a data set, performing data augmentation on the data set, and dividing the data set into a training set and a verification set according to a certain proportion;
s4, inputting the training set and the verification set into a target detection network based on deep learning for training to generate a detection model;
and S5, inputting the high-voltage cable connector polishing picture to be detected into the trained detection model to obtain all defect positions and defect types in the picture.
Wherein, in step S1, utilize the industry camera to gather the high tension cable joint picture after artifical the polishing, specifically include:
as shown in fig. 2, for the high-voltage cable structure, four cameras (fig. (1)) are used to shoot vertically to the high-voltage cable at four angles, up, down, left and right, each camera is kept at the same distance from the cable (fig. (2)), and the four cameras are spatially on the same plane, are perpendicular to the side surface (fig. (4)) of the cable, and are parallel to the top surface.
As shown in fig. 3, in step S2, after the obtained picture is preprocessed and background-segmented, the position and the type of the defect are manually marked on the segmented picture by using marking software, which specifically includes:
s201, preprocessing the picture:
firstly, due to the instability of ambient light, the acquired image may have the defects of uneven brightness, noise, deformation and the like, which affect the extraction of the network to the image characteristics, thereby reducing the network precision, so that the image needs to be preprocessed to improve the defects; the specific pretreatment method comprises the following steps:
(1) Correcting the uneven illumination of the picture
Specifically, an image is corrected by adopting an illumination unevenness image adaptive correction algorithm based on a two-dimensional gamma function. Firstly, light components are obtained by using multi-scale Gaussian filtering of Retinex, then brightness change is carried out on V (brightness) components of HSV (hue, saturation and value) space of an original image by using a two-dimensional Gamma function, and results are obtained, wherein the used Gaussian function is as follows:
wherein c is a scale factor; λ is a normalization constant, ensuring that the gaussian function G (x, y) satisfies a normalization condition, i.e., — > jg (x, y) dxdy =1, and performing convolution by using the gaussian function and the original image to obtain an estimated value of the illumination component, wherein the result is as follows:
wherein F (x, y) is an input image;i (x, y) is the estimated illumination component. Omega i Extracting a weight coefficient of the illumination component for the ith scale Gaussian function; i =1, 2.. N is the number of scales used, and is the accuracy of the balance calculation amount and the illumination component extraction, where N =3 is taken, i.e. the values of the illumination component are extracted by using gaussian functions of 3 different scales.
(2) Noise removal using gaussian filtering
Specifically, gaussian filtering is a linear smooth filtering, is suitable for eliminating gaussian noise, and is widely applied to a noise reduction process of digital image processing. Generally speaking, gaussian filtering is a process of performing weighted average on the whole image, and the value of each pixel point is obtained by performing weighted average on the value of each pixel point and other pixel values in the neighborhood.
The specific operation of gaussian filtering is: each pixel in the image is scanned using a template (or convolution, mask) and the weighted average gray value of the pixels in the neighborhood determined by the template is used to replace the value of the center pixel of the template. The weight of each pixel in its neighborhood is equal for mean filtering and block filtering. In the Gaussian filtering, the weights of convolution kernels or templates are in a Gaussian distribution form, namely the weight of a central point is increased, the weight far away from the central point is decreased, and the sum of different weights of each pixel value in a neighborhood is calculated on the basis.
The size of the convolution kernel is chosen according to the actual situation. Large convolution kernels work well for image noise processing, but can destroy-part of the image details, causing image distortion. The small convolution kernel has small damage degree to image details, but has undesirable noise removal effect, and the specific selection needs to be determined according to the quality of the current image and the effect of the image which is wanted to be acquired later.
S202, performing background segmentation on the picture:
secondly, due to environmental reasons, the shot picture contains environmental background information, which is not beneficial to subsequent detection, and therefore, the influence of irrelevant background on the target area needs to be eliminated. Further, as shown in fig. 2, when the camera takes images from four angles, the images may overlap (fig. (3)), and therefore, the obtained picture is subjected to background segmentation and overlapping region removal.
(1) Background removal using HSV color threshold based segmentation
And converting the image smoothed by the Gaussian filtering from RGB into HSV. H refers to hue, and colors such as white, red, green and the like which are seen in daily life are all described by H. The angle can be generally used for measuring H, and the value range is 0-360 degrees; s is the saturation degree, which expresses the purity of the color, the higher the S is, the purer the color is, the lower the S is, the gray color gradually changes, and a numerical value of 0-100 percent is taken; v is lightness representing the degree of brightness of the color, and for a light source color, the lightness value is related to the brightness of the illuminant; for object colors, this value is related to the transmittance or reflectance of the object, and typically ranges from 0% (black) to 100% (white). HSV works well for color thresholding, the color model of which is shown in fig. 4, and the formula for converting RGB to HSV is as follows:
C max ＝max(R',G',B')
Δ＝C max -C min
h, calculation:
s, calculating:
v, calculation:
V＝C max
after the noise of the image is removed through Gaussian filtering and HSV color space conversion, on the basis of an HSV color model, a proper color threshold (comprising an upper threshold and a lower threshold) is selected to distinguish the cable image from a background area.
(2) Width shrinking the image:
since the absolute distance between the camera and the cable is fixed ((2) in fig. 2), the overlapping area is also fixed ((3) in fig. 2), and the left and right sides of the picture obtained by shooting are respectively contracted by a distance of one half of the distance in (3) in fig. 2, so that repeated detection is avoided, the calculation amount is reduced, and the detection speed is improved.
S203, manual labeling:
and finally, labeling the cut data by using Labelme labeling software, and generating a data set in a COCO or VOC format according to the selected target detection network data format.
In step S3, the marked picture is cut to form a data set, the data set is augmented, and the data set is divided into a training set and a verification set according to a certain proportion, which specifically includes:
since the resolution of images captured by an industrial camera may be very high, such as 3072 × 2018, even after background removal, it is inconvenient to operate, and the computational requirements on a computer are very high if the images are directly sent to a network for training. The picture is split into 800 x 800 small pictures, taking into account the limited performance of the computers used.
Because the number of the high-voltage cable connectors which are manually polished is limited, and deep learning which depends on data volume is very unfavorable, a certain data amplification method is needed to be adopted to expand data so as to meet the number required by network training. Specifically, the data augmentation method comprises the following steps:
and carrying out data enhancement methods such as random image inversion, random brightness transformation, random scale scaling and the like on the graph. And dividing the augmented data sample set into a training sample set and a verification sample set according to a certain proportion. The usual proportions are 7:3,8:2,9:1, for small sample data sets, it is generally desirable that the training set ratio be as large as possible in order to achieve higher accuracy. The overall flow of this step is shown in FIG. 5.
In step S4, the training set and the verification set are input to a target detection network based on deep learning for training, and a detection model is generated, which specifically includes:
and inputting the distributed training set and the verification set into a target detection deep neural network for end-to-end learning training. Among deep learning networks that may be used are CenterNet, faster-RCNN, effectientNet, or DETR. Taking the centrnet as an example:
the common target detection algorithm usually uses the setting of a priori frame, that is, a great number of priori frames are set on a picture, the priori frame is adjusted by the prediction result of the network to obtain the prediction frame, the detection capability of the network is improved to a great extent by the priori frame, but the size of the object is limited. The Centernet adopts different methods, a target is taken as a point when a model is built, namely the central point of the target BBox, and a detector of the Centernet adopts key point estimation to find the central point and regresses to other target attributes.
The central VOC2012 data set is used as a pre-training model, and the method uses Resnet50 as the backbone Network, and uses a model trained on a Pascal VOC2012 data set as the pre-training model because the Hourglass Network, DLANet or Resnet used by the central is a backbone feature extraction Network, and because the Hourglass Network parameters used by the central are too large and have 19000W parameters and DLANet has no keras resources.
The training process is divided into a freezing training stage and a thawing training stage. The effect of freeze training is to freeze the weights of the general network part in order to place more resources on the network parameters of the later part of the training, which results in a significant improvement in both time and resource utilization. These frozen portions are then thawed after a period of later network parameter training, all together.
In this example, 50 rounds of freeze training, batch _ size set to 6, initial learning rate set to le-3, thaw training set to 250 rounds, batch _ size set to 2, initial learning rate set to 1e-4, and other hyper-parameter settings are as follows:
(1) Adopting an Adam optimizer, wherein the weight attenuation coefficient is 4e-4;
(2) Adjusting a learning strategy Step _ LR at equal intervals, wherein the Step _ size is adjusted to be 1, and the magnification gamma =0.93 is adjusted;
in the training process, verification and model storage are carried out in each period, and the model with the highest verification accuracy is compared and stored at the same time.
As shown in fig. 6, in step S5, inputting a polishing picture of the high-voltage cable connector to be detected into a trained detection model to obtain all defect positions and defect types in the picture, specifically including:
firstly, carrying out illumination enhancement on a picture to be detected, wherein the specific illumination enhancement method is an illumination nonuniformity image self-adaptive correction algorithm based on a two-dimensional gamma function.
And then, performing background segmentation on the enhanced picture, wherein the specific background segmentation method is based on an HSV color threshold segmentation algorithm.
Then, for the convenience of training, the picture without the background is split into small images of N × N, and the small images are sent to a trained target detection network for detection.
And finally, splicing the detected small pictures back to the original size, and outputting the image.
According to the high-voltage cable joint polishing defect detection method based on deep learning, the target detection method based on deep learning is applied to high-voltage cable joint polishing defect detection for the first time, and a set of complete and feasible detection method and operation flow are provided; the effectiveness and feasibility of the defect detection method based on vision are proved, and a new case is added for the defect detection method based on digital images in the industry.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the foregoing, the description is not to be taken in a limiting sense.
Claims (6)
1. A high-voltage cable joint polishing defect detection method based on deep learning is characterized by comprising the following steps:
s1, acquiring a picture of a high-voltage cable connector which is polished manually by using an industrial camera;
s2, after preprocessing and background segmentation are carried out on the obtained picture, the position and the type of the defect are manually marked on the segmented picture by utilizing marking software;
s3, cutting the marked picture to form a data set, performing data augmentation on the data set, and dividing the data set into a training set and a verification set according to a certain proportion;
s4, inputting the training set and the verification set into a target detection network based on deep learning for training to generate a detection model;
and S5, inputting the polished picture of the high-voltage cable joint to be detected into the trained detection model to obtain all defect positions and defect types in the picture.
2. The method for detecting polishing defects of a high-voltage cable joint based on deep learning of claim 1, wherein in the step S1, an industrial camera is used for acquiring a picture of the high-voltage cable joint after manual polishing, and the method specifically comprises the following steps:
the four industrial cameras are respectively positioned at four angles, namely, the upper angle, the lower angle, the left angle and the right angle and are perpendicular to the high-voltage cable to shoot, the distance between each industrial camera and the high-voltage cable is equal, the four industrial cameras are positioned on the same plane in space and are perpendicular to the side face and parallel to the top face of the high-voltage cable.
3. The deep learning-based high-voltage cable joint polishing defect detection method as claimed in claim 1, wherein in step S2, after preprocessing and background segmentation are performed on the obtained picture, the position and the type of the defect are manually marked on the segmented picture by using marking software, and specifically, the method comprises:
s201, preprocessing the picture: correcting the phenomenon of uneven illumination of the picture by adopting an illumination unevenness image self-adaptive correction algorithm based on a two-dimensional gamma function, and removing noise by adopting Gaussian filtering;
s202, performing background segmentation on the picture: removing the background by using an HSV color threshold segmentation based method, and then shrinking the width of the picture;
and S203, labeling the divided pictures by using Labelme labeling software, and generating COCO or VOC format storage according to the selected target detection network data format.
4. The method for detecting the polishing defect of the high-voltage cable joint based on the deep learning of claim 1, wherein in the step S3, the marked picture is cut to form a data set, the data set is subjected to data augmentation, and the data set is divided into a training set and a verification set according to a certain proportion, and specifically comprises:
cutting the marked picture into small pictures of 800 multiplied by 800;
random image turnover, random brightness transformation and random scale scaling are carried out on the data set, and data augmentation is achieved;
according to the weight ratio of 7: 3. 8:2 or 9:1, the data set is divided into a training set and a validation set.
5. The deep learning-based high-voltage cable joint grinding defect detection method according to claim 1, wherein in the step S4, the training set and the verification set are input to a deep learning-based target detection network for training to generate a detection model, specifically comprising:
adopting CenterNet as a target detection network, using Resnet50 as a backbone network, and using a model trained on a Pascal VOC2012 data set as a pre-training model;
the training process comprises a freezing training stage and a thawing training stage, 50 rounds of freezing training are performed, the batch _ size is set to be 6, the initial learning rate is set to be le-3, 250 rounds of thawing training are performed, the batch _ size is set to be 2, the initial learning rate is set to be 1e-4, an Adam optimizer is adopted, the weight attenuation coefficient is 4e-4, an equal-interval adjustment learning strategy Step _ LR is adopted, the adjustment interval Step _ size is 1, and the adjustment multiplying power gamma =0.93;
in the training process, verification and model storage are carried out in each period, and the model with the highest verification accuracy is compared and stored.
6. The deep learning-based high-voltage cable joint grinding defect detection method according to claim 1, wherein in the step S5, the high-voltage cable joint grinding picture to be detected is input into the trained detection model, so as to obtain all defect positions and defect types in the picture, and specifically comprises:
carrying out illumination enhancement on a polished picture of the high-voltage cable joint to be detected, wherein the specific illumination enhancement method is an illumination unevenness image self-adaptive correction algorithm based on a two-dimensional gamma function;
performing background segmentation on the enhanced picture, wherein the specific background segmentation method is based on an HSV color threshold segmentation algorithm;
splitting the picture without the background into small N × N pictures, and sending the small N pictures into a trained detection model for detection;
and splicing the detected small images back to the original size in sequence, and outputting the images.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211230402.1A CN115731166A (en) | 2022-10-08 | 2022-10-08 | High-voltage cable connector polishing defect detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211230402.1A CN115731166A (en) | 2022-10-08 | 2022-10-08 | High-voltage cable connector polishing defect detection method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115731166A true CN115731166A (en) | 2023-03-03 |
Family
ID=85293514
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211230402.1A Pending CN115731166A (en) | 2022-10-08 | 2022-10-08 | High-voltage cable connector polishing defect detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115731166A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115908988A (en) * | 2023-03-09 | 2023-04-04 | 苏州苏映视图像软件科技有限公司 | Defect detection model generation method, device, equipment and storage medium |
| CN117115153A (en) * | 2023-10-23 | 2023-11-24 | 威海坤科流量仪表股份有限公司 | Intelligent printed circuit board quality detection method based on visual assistance |
-
2022
- 2022-10-08 CN CN202211230402.1A patent/CN115731166A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115908988A (en) * | 2023-03-09 | 2023-04-04 | 苏州苏映视图像软件科技有限公司 | Defect detection model generation method, device, equipment and storage medium |
| CN117115153A (en) * | 2023-10-23 | 2023-11-24 | 威海坤科流量仪表股份有限公司 | Intelligent printed circuit board quality detection method based on visual assistance |
| CN117115153B (en) * | 2023-10-23 | 2024-02-02 | 威海坤科流量仪表股份有限公司 | Intelligent printed circuit board quality detection method based on visual assistance |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Adaptive image enhancement method for correcting low-illumination images | |
| Celik | Spatial mutual information and PageRank-based contrast enhancement and quality-aware relative contrast measure | |
| Wang et al. | Simple low-light image enhancement based on Weber–Fechner law in logarithmic space | |
| CN115731166A (en) | High-voltage cable connector polishing defect detection method based on deep learning | |
| JP4452497B2 (en) | Method and system for modifying a digital image in consideration of noise | |
| US11238301B2 (en) | Computer-implemented method of detecting foreign object on background object in an image, apparatus for detecting foreign object on background object in an image, and computer-program product | |
| CN108182671B (en) | Single image defogging method based on sky area identification | |
| CN111882555B (en) | Deep learning-based netting detection method, device, equipment and storage medium | |
| CN117274113B (en) | Broken silicon wafer cleaning effect visual detection method based on image enhancement | |
| Steffens et al. | Cnn based image restoration: Adjusting ill-exposed srgb images in post-processing | |
| CN116228780A (en) | Silicon wafer defect detection method and system based on computer vision | |
| CN113569708A (en) | Living body recognition method, living body recognition device, electronic apparatus, and storage medium | |
| CN111047618B (en) | Multi-scale-based non-reference screen content image quality evaluation method | |
| CN113971681A (en) | Edge detection method for belt conveyor in complex environment | |
| Kumar et al. | Haze elimination model-based color saturation adjustment with contrast correction | |
| CN112991329A (en) | Image shadow detection and elimination method based on GAN | |
| CN112365425A (en) | Low-illumination image enhancement method and system | |
| Khan et al. | Shadow removal from digital images using multi-channel binarization and shadow matting | |
| CN115908967A (en) | Petrochemical device pipeline data sample balancing method based on cyclic generation countermeasure network | |
| CN114565563A (en) | Color steel plate surface abnormity detection method based on artificial intelligence | |
| Zhou et al. | Saliency preserving decolorization | |
| Chen et al. | GADO-Net: an improved AOD-Net single image dehazing algorithm | |
| Zhao et al. | Nonuniform illumination correction for underwater images through a pseudo-siamese network | |
| CN112116580B (en) | Detection method, system and equipment for camera support | |
| Niu et al. | An Adaptive Recovering Algorithm for The Color Bar Code Image Based on Gamma Transforms |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |